Music genre classification based on auditory image, spectral and acoustic features

Music genre is one of the conventional ways to describe music content, and also is one of the important labels of music information retrieval. Therefore, the effective and precise music genre classification method becomes an urgent need for realizing automatic organization of large music archives. Inspired by the fact that humans have a better automatic recognizing music genre ability, which may attribute to our auditory system, even for the participants with little musical literacy. In this paper, a novel classification framework incorporating the auditory image feature with traditional acoustic features and spectral feature is proposed to improve the classification accuracy. In detail, auditory image feature is extracted based on the auditory image model which simulates the auditory system of the human ear and has also been successfully used in other fields apart from music genre classification to our best knowledge. Moreover, the logarithmic frequency spectrogram rather than linear is adopted to extract the spectral feature to capture the information about the low-frequency part adequately. These above two features and the traditional acoustic feature are evaluated, compared, respectively, and fused finally based on the GTZAN, GTZAN-NEW, ISMIR2004 and Homburg datasets. Experimental results show that the proposed method owns the higher classification accuracy and the better stability than many state-of-the-art classification methods.

Automated summary

The paper discusses a new method to improve the accuracy of music genre classification, combining auditory image features, traditional acoustic features, and spectral features. Experimental results demonstrate that the proposed method outperforms many state-of-the-art classification methods in terms of classification accuracy and stability.